Trajectories of premorbid healthcare contact days in veterans with pancreatic cancer.

706 Background: Early detection of pancreatic cancer is a key clinical priority. Although prior work has evaluated machine learning models to identify at-risk patients for pancreatic cancer screening, this is limited by complexity and cost. "Contact days’’— days spent receiving healthcare outside the home for any reason or length of time—have previously been used to measure the ‘’time toxicity’’ of cancer care post-diagnosis. We sought to characterize trajectories of premorbid contact days, alongside standardized risk scores and healthcare costs to explore their use as potential markers for early pancreatic cancer diagnosis. Methods: We identified decedents with pancreatic adenocarcinoma diagnosed between 2012 and 2022 who had at least one primary care visit in the Veterans Affairs (VA) system in each of the two years prior to diagnosis using the VA Clinical Cancer Registry (VACCR) and the Corporate Data Warehouse (CDW). We grouped patients into two cohorts: (1) surgical (receipt of curative intent surgery), and (2) non-surgical and collected sociodemographic and clinical characteristics. During a 2-year premorbid period we calculated monthly contact days (24 data points) and plotted them as percentages (e.g., 3 contact days in a month=10%) and visualized their trajectory using cubic smoothing splines. We plotted trajectories for each sociodemographic and clinical subgroup and visually evaluated differences. We similarly plotted monthly Care Assessment Need (CAN) scores, a validated VA risk score identifying patients at high risk of hospitalization/death, and monthly VA healthcare costs. Results: We included 8,654 patients (649 surgical - median age 69 years, 74% white, 78% stage 1-2 and 8,005 non-surgical - median age 71 years, 72% white, 21% stage 1-2). For both groups, trajectories of contact days were similar. Baseline contact days (-24 to -12 months) were 7-8% (≈2 days/month). Contact days started to rise on average 9 months prior to diagnosis at an approximate rate of 1% per month, reaching 15-20% at diagnosis (≈5-6 days/month). In the surgical group, patients with a Charlson comorbidity score of ≥2, urban residence, and Black race had higher baseline contact days. Those <65 years and rural residence had a steeper slope of rise in contact days. CAN score trajectories closely mirrored contact day trajectories in both groups. Cost trajectories were similar in both groups (≈$1000/month at baseline), rose gradually to ≈$2000/month by -3 months, and then rose sharply to >$6000/month 2 months before diagnosis. Conclusions: In this decade-long national study of Veterans diagnosed with pancreatic cancer, we found that contact days and CAN scores, but not costs, start to rise 9 months prior to diagnosis. These data provide the first evidence supporting their further evaluation in pancreatic cancer risk models which may aid early diagnosis. The similar trajectories in patients who go on to receive curative-intent surgery or not may imply they may be markers for pancreas cancer diagnosis but perhaps not outcomes.